Evolution Strategies

evolve(e::AbstractEvolution, population)

Return a new population evolved by e.

New population may or may not contain same individuals as before.

evolve(e::AbstractEvolution, f::AbstractFitness, p::Population)

Return a new population fitted by f and evolved by e.

This is done by first replacing each member of p with a FittedCandidate with fitness computed by f.

Then evolve population into a new population using e. New population may or may not contain same individuals as before.

NoOpEvolution <: AbstractEvolution
NoOpEvolution()

Does not evolve the given population.

EliteSelection <: AbstractEvolution
EliteSelection(nselect::Integer, evo=NoOpEvolution())

Selects the nselect highest fitness candidates to be passed on to evo.

SusSelection <: AbstractEvolution
SusSelection(nselect, evo, rng=rng_default)

Selects candidates for further evolution using stochastic universal sampling.

TournamentSelection <: AbstractEvolution
TournamentSelection(nselect, k, p::Real, evo, rng=rng_default)

Selects candidates for further evolution using tournament selection.

Holds nselect tournaments with one winner each where each tournament has k random candidates from the given population.

Winner of a tournament is selected as the candidate with highest fitness with a probability p, second highest fitness with a probability p(p-1), third highest fitness with a probability of p((p-1)^2) and so on.

CombinedEvolution <: AbstractEvolution
CombinedEvolution(evos::AbstractArray)
CombinedEvolution(evos...)

Combines the evolved populations from several AbstractEvolutions into one population.

EvolutionChain <: AbstractEvolution
EvolutionChain(evos::AbstractArray)
EvolutionChain(evos...)

Chains multiple AbstractEvolutions in a sequence so that output from the first is input to the next and so on.

PairCandidates <: AbstractEvolution
PairCandidates(evo::AbstractEvolution)

Creates pairs of candidates in a population and calls evolve(evo, pairs) where pairs is the array of pairs.

ShuffleCandidates
ShuffleCandidates()
ShuffleCandidates(rng)

Shuffles the population using rng.

Useful with PairCandidates in case the prior selection does not shuffle the population.

EvolveCandidates <: AbstractEvolution
EvolveCandidates(fun)

Applies fun(c) for each candidate c in a given population.

Useful with MapCandidate.

AfterEvolution <: AbstractEvolution
AfterEvolution(evo, fun)

Return fun(newpop) where newpop = evolve(e.evo, pop) where pop is original population to evolve.